jaybee123/xente-fraud-api repository overview

⁠Enterprise Real-Time Fraud Detection API (Xente/Fintech)

⁠Overview

This image provides a production-ready, containerized Machine Learning REST API for detecting fraudulent financial transactions in real-time.

Built using FastAPI and an optimized XGBoost model trained on the Xente Fintech dataset (Mobile Money/E-commerce), this microservice evaluates incoming transaction payloads, applies custom Scikit-Learn behavioral feature engineering, and returns a probability score and fraud classification in milliseconds.

The underlying model was specifically engineered to handle high-cardinality categorical data and extreme class imbalance (AUPRC: 0.842, False Positive Rate: 0.094%), ensuring high recall with minimal friction to legitimate customers.

⁠🛠️ Tech Stack Inside the Container

• Serving Layer: FastAPI, Uvicorn, Pydantic (Strict Schema Validation)
• Machine Learning: XGBoost (v2.0+), Scikit-Learn (Custom Pipelines)
• MLOps: MLflow (Model unpickling, Signature enforcement)
• Base Image: python:3.10-slim

⁠🚀 Quick Start

1. Pull the image:

docker pull jaybee123/xente-fraud-api:latest

2. Run the container:

docker run -d --name fraud-api -p 8000:8000 jaybee123/xente-fraud-api:latest

3. Access the API: Once running, the interactive Swagger UI documentation is automatically available at: 👉 http://localhost:8000/docs⁠

⁠📡 API Usage & Endpoints

⁠GET /test

A built-in testing endpoint that runs a dummy transaction through the loaded MLflow model to verify inference integrity.

⁠POST /predict

The primary inference endpoint. It expects a JSON payload containing the raw transaction parameters. The internal Scikit-Learn transformer automatically extracts temporal and behavioral velocity features on the fly.

Example cURL Request:

curl -X 'POST' \
  'http://localhost:8000/predict' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "TransactionId": "TransactionId_51888",
  "CustomerId": "CustomerId_1",
  "ProviderId": "ProviderId_4",
  "ProductId": "ProductId_10",
  "ProductCategory": "airtime",
  "ChannelId": "ChannelId_2",
  "Amount": -10000.0,
  "Value": 10000.0,
  "TransactionStartTime": "2018-11-21T16:49:14Z",
  "PricingStrategy": 4
}'

Example JSON Response:

{
  "transaction_id": "TransactionId_51888",
  "Product_Category": "airtime",
  "fraudResult": false,
  "fraud_probability": 0.00004561
}

⁠⚙️ Configuration

The container exposes port 8000 by default. It natively contains the compiled MLflow model artifacts and the custom Scikit-Learn transformer class required for deserialization.

⁠🔗 Source Code & Documentation

This container is built and pushed automatically via GitHub Actions CI/CD pipelines. For the complete MLOps architecture, exploratory data analysis, and model training scripts, please visit the source repository:

• GitHub Repository: https://github.com/JUBRIL-ADEBOGUN/Fraud-Detection/⁠
• Dataset Context: Zindi Xente Fraud Detection Challenge